Every dataset
Pre-auditedleakage · duplicates · balance · de-id
Medical AI is only as reliable as the images used to train it.
ScaleiQ Image Bank provides automated image auditing, dataset validation, governance, provenance tracking, and readiness scoring in a single secure platform. By identifying hidden risks before training begins, ScaleiQ helps researchers accelerate development, improve reproducibility, and build more trustworthy AI models.
Sign up for our beta program.
Today that value is scattered across a general data-science hub, a model-and-dataset community, and siloed clinical biobanks — none of which tell you whether the images are actually clean. ScaleiQ brings the three together and adds the layer they're missing.
Consented, de-identified, provenance-tracked medical images with HIPAA/GDPR-aware governance — data you can actually put in a study.
Browse and search curated datasets by modality, disease, and label. Labeled classes, leakage-safe train/val/test splits, and a growing contributor community.
Every dataset ships with a card, image embeddings and similarity search, versioned exports, and API access — ready to pull into a training pipeline.
The reason you can't just grab a public medical-imaging set and train on it: you don't know if the test set leaks into training, whether the same eye appears fifty times, if one class dwarfs the rest, or whether a patient's name is still burned into a corner of the image. ScaleiQ runs the same VisionStream audit on every dataset before it's listed, and publishes the findings as a datasheet anyone can read.
OCT, fundus and retina are audited end to end today. DICOM and MRI are accepted, de-identified and catalogued now, and counted honestly as “queued” in the report rather than faked — full audit is on the roadmap.
Open a study with clinical records, or drop a labeled dataset — folders become classes and splits automatically.
DICOM tags scrubbed, burned-in pixel text OCR-redacted, head-MRI defaced, modality cross-checked — at ingest.
VisionStream flags leakage, duplicates, outliers and imbalance, then returns a readiness score and a datasheet.
Share a dataset card with your license and terms. Requesters ask; you approve or decline. Images stay private until you say so.
Browse curated cards, request access, and export a ready-to-train set — images foldered by split and class, plus the manifest.
Unlike an open biobank or a data broker, a published card is an image-free datasheet — composition, balance, splits, de-identification and readiness — so a dataset can be evaluated without anyone touching the images. Access happens only when the owner approves a request, and every grant is scoped and time-limited.
Need to upload more than 50 images? Send an email to our support team at support@scaleiqai.com.
Single users aren't the whole story. Institutions need private, compliant space for their own imaging — plus the ability to license, discover, and collaborate across the wider bank. The Institution plan gives your organization its own governed environment on top of the curated bank, provisioned to your requirements.
An isolated tenant for your organization — your datasets stored separately, with optional customer-managed encryption and data residency.
Team seats and roles, SSO, scoped and time-limited grants, and audit logs of who accessed what and when.
Programmatic search and retrieval, image embeddings and similarity search, and governed exports that fit into your ML pipelines.
HIPAA/GDPR-aware de-identification, provenance tracking, and auditability — the evidence enterprise data-governance reviews ask for.